Cortex A53 is good but not enough. Cortex A7 is still a better core in its time.
P/s: It's nice to see Andrei back, and especially when he said about "Snapdragon 820 vs Exynos 8890". Andrei, we're still waiting for Galaxy S7 review part II.
Ok so I know this has nothing to do with this article. And I know I will get flack for this. Nor am I sating that this article is of lesser importance not at all!! In fact I think this is a very important article. I'm usually the last to complain but... WHERE is the GTX 1080 review? I've been a long time Anandtech reader and am a loyal fan but am concerned about misses like this. If you guys have discovered something that requires extra time I get that but then there is still the failure of letting us know this fact. Either way this is an excessively long time for a GPU review. Luckily for myself I am some months away from building my next system so I can wait. Although my current go to is probably a 1080. However a lot of early adopters would have been relying on reviews out of the gate. And however you view early adopters, weather you think they are wasting their money or not, they make up a very significant percentage of those who would benefit from a timely review. Let's face it, already by now, there are plenty of other very detailed and accurate reviews out there. I understand wanting to be detailed oriented. However others are in the same game and are often just as passionate. Sorry I don't mean to dis anyone as I am a long time Anandtech reader but I am concerned.
if you cant get the review out by the time the cards go on sale then something is wrong. That's the whole point of a review right? Especially when you guys were given samples 2 weeks ahead of time.
Work on the review as a team like you do on some of the other stuff with 3-4 authors. This should help you get it out quicker, and the editor can tidy things up
Please pick up your toys and carefully place them back in your respective prams.
If you hadn't noticed, AT never had major review out for release (viz iPhone). The plain facts are that quality research and reviews takes time, and manufacturers embargos don't give enough time for in depth work.
If you don't believe me then compare the quality and depth of AT reviews with the ones which do make it out when embargo lifts. I didn't see any of the early iPhone 5S reviews discussing the details of the Cyclone architecture for example.
If you're going to throw a childish tantrum, please aim it at OEMs for giving limited embargo time. Alternately just wait a few weeks for an in depth review before buying, if it matters that much to you.
For example I've had my GTX 970 for over a year and a half now. Looking back I don't think waiting a few weeks would changed the last 18 months of gaming pleasure/caused me to grow a third arm/ended the world etc.
The white paper on Pascal has been out for over a month, that has most of the in-depth stuff about the architecture. Yes anandtech does do in-depth testing, and it is of high quality, but so do PCPer, Toms, and Guru3D. All of which had their reviews posted when the embargo lifted. Hell, they even have the GTX 1070 review out. I'm not comparing them to "the verge" or "ars", what i'm saying is, Anand is no longer the only one posting highly technical reviews, and the competition is catching up big time this year.
This is professional level of incompetence tbf. You guys have had the GTX 1080 units for close to 3 weeks. Possibly the bigger tech reveal of the last 6 months when it comes to gaming (not counting any near future cpu releases too as those are lame ~5% improvements) and when 99% of the tech sites have their comprehensive reviews up and you don't, something is wrong as of late.
Normally takes them more than three weeks to get their iPhone reviews up and that's the biggest tech reveal if the year.
TBH now that Computex is on I doubt you'll see anything for a bit. It's the biggest PC shoe of the year (so I guess you will want prompt coverage) and the trip from the US to Asia is a long schlep.
Plus I'm sure the AT writers will want some down time while you get back - while you may prefer your tech news to be 24/7 in reality even tech writers have wives, kids and he right to have at least some weekend's off to enjoy a normal family life...
Im more interested in an arm highly efficient processor, than in a 16nm expensive power hungry behemoth like nvidia 1080 gtx. Human overpopulation is enough so to also have the same overreproducting guys asking for powerhouses. DId you heard about coal, fracking, radiation waves from fukushima in ocean.
What's the big deal? The Founder's cards with the $100 early adopter tax sold out immediately anyway, and the more interesting cards at reasonable prices won't ship for a while.
" and given the stated release schedule and looking back at last year’s SoCs with A72 cores, it shouldn't be too hard to guess which device and chipset might be able to fit this timeline. "
I really don't follow the mobile world very closely these days... Anyone wanna spell that guess out for me?
HiSilicon and the next Mate would perfectly fit the quoted release schedule. I didn't want to put spell this out in the article since it is after all only a guess.
As much as I'd like to see that, it seems unlikely... The next Nexus is likely already being finalized (if it's gonna go on sale in 5-6 months), work on it would've started a while ago. Last year's just used 810 didn't it?
Yes. The new Nexus is almost guaranteed to use 820, especially if it's built by HTC (as the rumours say it is). If it is made by Huawei however, things could get interesting.
Mediatek was first with A72 in the Amazon TV box and in mobile Huawei was first. Mediatek has said that their high end SoC (likely called Helio x30) might ship this year but devices might have to wait for next year and Huawei is rumored to have something new on 16FF not to ofar ahead- 10nm isn't available this year, that's more like second half of 2017 at best for phones in retail.
Yeah it was but the question wasn't process specific. Ofc with A72 Qualcomm was fast too but hard to predict when and if they use A73. Spreadtrum might get more aggressive as they try to gain share above the low end but no idea what they are planning beyond their soon to arrive 8xA53 + T880MP4 on 16FFC.
On the slide comparing 8xA53 with a hexa core with 2xA73, any clue what they use to measure perf and more importantly in what power budget the perf claims are? Not that we'll see much of it on 28nm when everybody went FinFET already with A53 based SoCs.
Anyway , this sounds interesting even if CPU perf is somewhat enough at this point. More curious about G71, what they'll do with the Apical and if they make something for machine learning beyond Apical and/or GPU.
Decreasing the instruction decode width from 3/cycle to 2 to save power is interesting. When you look at the x86 vs RISC debate, everyone tells you complex instruction decoders are a small cost compared to the rest of the core, which is true in terms of area, but no mention about power?
So for A73, did ARM reduce the decode width because it was rarely being utilized, or because fundamentally it's power hungry, or both? This seems to suggest why there are no x86 phones. If instruction decode power is a problem for RISC ARM, then it should be even a bigger problem for CISC x86.
IIRC, the original K7 only manages to sustain execution on just one decoder out of three, and successive x86 uarches has made decoders more and more capable on per decoder basis, while almost all the x86 IPC gains seems to come from everything except the decoders...I can see the case why ARM moved from 3 to 2 decoders.
> For example, the A15, A57, A72 all belong to the Austin family of microarchitectures, and as one would have guessed from the name, this is because they originated from ARM's Austin CPU design centre.
The A5, A7 and A53 belong to the Cambridge family while the Cortex A12, A17 and today's new A73 belong to the Sophia family, owning its name to the small city of Sophia-Antipolis which houses one of Europe's largest technology parks as well as ARM's French CPU design centre.
So the American ARM CPU cores suck, while the European ones are pretty great. Got it. We should keep this in mind when ARM announces new cores. Stay away from the American ones, adopt the European ones.
The American designs use more power and are larger. The Cambridge designs try to use the least power and are small (I expect the A35 is also from here?) The French designs are larger, but low power.
Each site's designs seem to take about two years, and they're staggered.
The biggest advantage is that if one of the sites really screws up a generation of CPU, it appears the other site will have them covered. If A73 was a turd, nobody would have batted an eyelid at A72 for another year, for example.
"...both microarchitectures have trouble in terms of differentiating themselves in terms of performance and power compared to ARM’s own current designs..."
You're kidding, right? *cough single threaded performance *cough....
I'll cover the topic in more detail in the separate review but neither Qualcomm nor Samsung display an advantage in single-thread performance over current ARM microarchitectures.
Chances are Qualcomm will go FOWLP soon and that's why they want custom A73. On 10nm the die will be tiny and plenty of room to add the DRAM in the same package.
A72s @ 28nm are already scoring ~1700 Geekbench ST. If A73 on the 28nm adds a conservative 20% puts it in Kyro/Mongoose territory. 14nm A73 would be even stronger still with inherent finfet power efficiency boosts.
Not hard to see why custom ARM camp should be really worried, because ARM own designs have the flexibility in process nodes while performance is rapidly catching up with custom.
2016 / 2017 are going to be interesting again in the mobile SoC space.
We have the fully custom SoCs from the likes of Qualcomm (Kryo CPU, and Adreno GPU).
The semi-custom SoCs from the likes of Samsung (Mongoose big cores + Cortex-A53 LITTLE cores, and Mali GPU) and Apple (Twister CPU, and PowerVR GPU).
The pure ARM SoCs from the likes of MediaTek, Rockchip, Huewai (sp?), and similar (Cortex-A72/A73 + Cortex-A53/A35 and Mali GPUs).
All on multiple processes (28nm, 16nm, 14nm, high-performance, low-power, FinFETT, etc).
Now we just need someone to combine a high-performance, high-efficiency SoC with a high-efficiency radio and screen (that's not too big) and stick a huge battery inside (4000 mAh should be the new normal; anything under 3000 mAh should be avoided). The era of 2-days-between-charges could be upon us ... so long as the OEMs don't squander the battery savings by making the phones even thinner than they are now!!
2015 may have been boring. But it's certainly not anymore!
I also don't think this will be the last of ARM's big core microprocessor family (A15, A57, A72). As the A73 is purely focused on the consumer market, and that's a good thing. However I don't think ARM is giving up on industrial usage and its dream of breaking into the server market. I expect there will be a successor to the A72, and now that it doesn't have to pretend to want to be used in a mobile phone, it can go bigger and wider. It will probably be named something like Cortex A82 or A92 or something, and it could be an absolute beast in Chromebooks, laptops, and potentially maybe even some micro servers. It would be great competition to Intel's Core m, at a fraction of the price.
Fun times to be a microprocessor nerd, fun times!
Thanks Andrei for your great reporting. You are making me almost not miss Anand anymore :-)
This seems like an awesome replacement for the A72 but why would you need an A53 now. If two A73s can do more with the same space and power budget as 4 A53s why use the core? Honestly I'm kinda mad this was not the core that was initially pushed as the 64 bit chip. ARM was in a great place with the A17 then they pushed forward with a core that was too big hot and power hungry for mobile devices. Honestly I would love to see what a manufacturer can do with six of these bad boys on a soc. May be overkill but it could be fun.
A53 perf/watt is much better than the A72. Take a look at the Kirin 950 results. You can more or less double your efficiency with an A53 vs downclocking an A72.
He was referencing the A73 not the A72. The A73 has nearly the same efficiency as A53 based on the graph, certainly 2 A73's would use less power than 4 A53's, if the graph is to be believed.
It has nothing to do with Apple, it is completely obvious jumping from 32-bit to 64-bit means one of the processor generations has to take one for the team.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
63 Comments
Back to Article
revanchrist - Sunday, May 29, 2016 - link
So we got A73 big core now, but still no little core A53 succesor. A35 is hardly a successor to A53.Ariknowsbest - Monday, May 30, 2016 - link
The A53 is still good, and likely harder to improve in a meaningful way as it's in-order design. The A35 is the successor to the A5.But it's nice with a competitive big core
hung2900 - Monday, May 30, 2016 - link
Cortex A53 is good but not enough. Cortex A7 is still a better core in its time.P/s: It's nice to see Andrei back, and especially when he said about "Snapdragon 820 vs Exynos 8890". Andrei, we're still waiting for Galaxy S7 review part II.
Krysto - Monday, May 30, 2016 - link
Well, that's why the A35 is more of a direct (64-bit) successor to the A7.revanchrist - Sunday, May 29, 2016 - link
And there's also a new GPU called Mali G71 unveiled by ARM today. Looks quite impressive, MP16 for 4W power consumption.https://benchlife.info/arm-announce-mali-g71-with-...
Locut0s - Sunday, May 29, 2016 - link
Ok so I know this has nothing to do with this article. And I know I will get flack for this. Nor am I sating that this article is of lesser importance not at all!! In fact I think this is a very important article. I'm usually the last to complain but... WHERE is the GTX 1080 review? I've been a long time Anandtech reader and am a loyal fan but am concerned about misses like this. If you guys have discovered something that requires extra time I get that but then there is still the failure of letting us know this fact. Either way this is an excessively long time for a GPU review. Luckily for myself I am some months away from building my next system so I can wait. Although my current go to is probably a 1080. However a lot of early adopters would have been relying on reviews out of the gate. And however you view early adopters, weather you think they are wasting their money or not, they make up a very significant percentage of those who would benefit from a timely review. Let's face it, already by now, there are plenty of other very detailed and accurate reviews out there. I understand wanting to be detailed oriented. However others are in the same game and are often just as passionate. Sorry I don't mean to dis anyone as I am a long time Anandtech reader but I am concerned.Ryan Smith - Sunday, May 29, 2016 - link
"WHERE is the GTX 1080 review?"Take a deep breath. It's nearly finished.=)
We're waylaid with Compuex for the moment, a couple of other things that need to get posted first.
Morawka - Monday, May 30, 2016 - link
if you cant get the review out by the time the cards go on sale then something is wrong. That's the whole point of a review right? Especially when you guys were given samples 2 weeks ahead of time.Work on the review as a team like you do on some of the other stuff with 3-4 authors. This should help you get it out quicker, and the editor can tidy things up
Jon Tseng - Monday, May 30, 2016 - link
Hi Morawka/Locut0sPlease pick up your toys and carefully place them back in your respective prams.
If you hadn't noticed, AT never had major review out for release (viz iPhone). The plain facts are that quality research and reviews takes time, and manufacturers embargos don't give enough time for in depth work.
If you don't believe me then compare the quality and depth of AT reviews with the ones which do make it out when embargo lifts. I didn't see any of the early iPhone 5S reviews discussing the details of the Cyclone architecture for example.
If you're going to throw a childish tantrum, please aim it at OEMs for giving limited embargo time. Alternately just wait a few weeks for an in depth review before buying, if it matters that much to you.
For example I've had my GTX 970 for over a year and a half now. Looking back I don't think waiting a few weeks would changed the last 18 months of gaming pleasure/caused me to grow a third arm/ended the world etc.
Now back to our normal programming...
J
Morawka - Monday, May 30, 2016 - link
The white paper on Pascal has been out for over a month, that has most of the in-depth stuff about the architecture. Yes anandtech does do in-depth testing, and it is of high quality, but so do PCPer, Toms, and Guru3D. All of which had their reviews posted when the embargo lifted. Hell, they even have the GTX 1070 review out. I'm not comparing them to "the verge" or "ars", what i'm saying is, Anand is no longer the only one posting highly technical reviews, and the competition is catching up big time this year.tuxRoller - Monday, May 30, 2016 - link
If those sites have such high quality, in-depth reviews why do you need to have a fourth?bji - Monday, May 30, 2016 - link
His comments were hardly a childish tantrum. Your out of proportion melodramatic response, however, would seem to qualify.milkywayer - Monday, May 30, 2016 - link
This is professional level of incompetence tbf. You guys have had the GTX 1080 units for close to 3 weeks. Possibly the bigger tech reveal of the last 6 months when it comes to gaming (not counting any near future cpu releases too as those are lame ~5% improvements) and when 99% of the tech sites have their comprehensive reviews up and you don't, something is wrong as of late.Jon Tseng - Monday, May 30, 2016 - link
Normally takes them more than three weeks to get their iPhone reviews up and that's the biggest tech reveal if the year.TBH now that Computex is on I doubt you'll see anything for a bit. It's the biggest PC shoe of the year (so I guess you will want prompt coverage) and the trip from the US to Asia is a long schlep.
Plus I'm sure the AT writers will want some down time while you get back - while you may prefer your tech news to be 24/7 in reality even tech writers have wives, kids and he right to have at least some weekend's off to enjoy a normal family life...
Demiurge - Monday, May 30, 2016 - link
Uh... wow, someone didn't get their nap today!Sometimes, good things come to those who wait...
Alexey291 - Monday, May 30, 2016 - link
Eh let's face it... AT is now a forum first and review website second.Pretty sure most of their traffic goes straight to the forums without even stopping at the main site.
Evidently the review side of the business is getting neglected somewhat.
osxandwindows - Sunday, May 29, 2016 - link
Agreed.I have seen reviews of the 1080 that are just as detailed or more then anand tech's reviews.
Jon Tseng - Monday, May 30, 2016 - link
I've actually seen 1080 reviews a lot worse than the performance preview Anandtech DID get out for embargo.Take Polygon for example, whjch burned the 1080 for being overkill if you want to use it for 1080p gaming:
http://www.polygon.com/2016/5/17/11689352/nvidia-g...
That's what you get if you rush things...
humanentity - Monday, May 30, 2016 - link
Im more interested in an arm highly efficient processor, than in a 16nm expensive power hungry behemoth like nvidia 1080 gtx. Human overpopulation is enough so to also have the same overreproducting guys asking for powerhouses. DId you heard about coal, fracking, radiation waves from fukushima in ocean.Alexey291 - Monday, May 30, 2016 - link
That's nice and I'm not interested in another paper release. So what makes you better than me?Michael Bay - Tuesday, May 31, 2016 - link
Stop sending unearned food to Africa and there will be no overpopulation in the record timeframe.Then you can kvetch about radiation waves.
Impulses - Monday, May 30, 2016 - link
What's the big deal? The Founder's cards with the $100 early adopter tax sold out immediately anyway, and the more interesting cards at reasonable prices won't ship for a while.zeeBomb - Monday, May 30, 2016 - link
Snapdragon 830 bound! Just 10%-30% efficiency on various areas...nice!Impulses - Monday, May 30, 2016 - link
" and given the stated release schedule and looking back at last year’s SoCs with A72 cores, it shouldn't be too hard to guess which device and chipset might be able to fit this timeline. "I really don't follow the mobile world very closely these days... Anyone wanna spell that guess out for me?
Andrei Frumusanu - Monday, May 30, 2016 - link
http://www.anandtech.com/show/9762/hisilicon-annou...http://www.anandtech.com/show/9878/the-huawei-mate...
HiSilicon and the next Mate would perfectly fit the quoted release schedule. I didn't want to put spell this out in the article since it is after all only a guess.
Impulses - Monday, May 30, 2016 - link
Thanks, very interesting.JamesU - Monday, May 30, 2016 - link
I suppose they're talking about a new Nexus with Snapdragon 830 but not sure.Impulses - Monday, May 30, 2016 - link
As much as I'd like to see that, it seems unlikely... The next Nexus is likely already being finalized (if it's gonna go on sale in 5-6 months), work on it would've started a while ago. Last year's just used 810 didn't it?aryonoco - Monday, May 30, 2016 - link
Yes. The new Nexus is almost guaranteed to use 820, especially if it's built by HTC (as the rumours say it is). If it is made by Huawei however, things could get interesting.jjj - Monday, May 30, 2016 - link
Mediatek was first with A72 in the Amazon TV box and in mobile Huawei was first.Mediatek has said that their high end SoC (likely called Helio x30) might ship this year but devices might have to wait for next year and Huawei is rumored to have something new on 16FF not to ofar ahead- 10nm isn't available this year, that's more like second half of 2017 at best for phones in retail.
Ian Cutress - Monday, May 30, 2016 - link
Wasn't the Amazon A72 box on 28nm or 20nm?Excors - Monday, May 30, 2016 - link
Yes, it's using an MT8173 (2xA72+2xA53), which is 28nm HPM.jjj - Monday, May 30, 2016 - link
Yeah it was but the question wasn't process specific.Ofc with A72 Qualcomm was fast too but hard to predict when and if they use A73. Spreadtrum might get more aggressive as they try to gain share above the low end but no idea what they are planning beyond their soon to arrive 8xA53 + T880MP4 on 16FFC.
jjj - Monday, May 30, 2016 - link
On the slide comparing 8xA53 with a hexa core with 2xA73, any clue what they use to measure perf and more importantly in what power budget the perf claims are? Not that we'll see much of it on 28nm when everybody went FinFET already with A53 based SoCs.Anyway , this sounds interesting even if CPU perf is somewhat enough at this point. More curious about G71, what they'll do with the Apical and if they make something for machine learning beyond Apical and/or GPU.
djayjp - Monday, May 30, 2016 - link
The bar for memory on the last page, first chart, clearly isn't 15% ha. I know it's ARM, not AT though.Tralalak - Monday, May 30, 2016 - link
Ad Cortex A17:With good graphics (GPU) is out-of-order Cortex A17 still very powerfull and energy efficient.
Zhaoxin ZX 5880 OctaCore ARM big.LITTLE SoC. 4x Cortex A17 up to 2.2GHz (2MB L2 cache) + 4x Cortex A7 1.4GHz (512kB L2 cache) + Elite2000 iGPU with Unified shader Arch with 256 stream processor
link: http://browser.primatelabs.com/geekbench3/search?u...
link: http://en.zhaoxin.com/site/product_detail/139
Zhaxoin ZX-2000: 4x Cortex A17 up to 1.6GHz
link: http://www.cfan.com.cn/2016/0414/125732.shtml
karthik.hegde - Monday, May 30, 2016 - link
Guess which CPU SD830 will have!!phoenix_rizzen - Monday, May 30, 2016 - link
Qualcomm Kryo would be my guess. Why would they go back on their own custom CPU cores now?Eden-K121D - Monday, May 30, 2016 - link
Well they hinted so that they would specifically use other components of an Soc to differentiate themselves other than the CPUghanz - Monday, May 30, 2016 - link
Looking forward to the upcoming deep dive article (with Snapdragon 820’s Kryo and Exynos 8890’s Mongoose cores). :)ToTTenTranz - Monday, May 30, 2016 - link
Are you guys going to cover the Mali G71 GPU architecture?Andrei Frumusanu - Monday, May 30, 2016 - link
Ryan is currently putting the finishing touches on that piece, it should go up later today.Ryan Smith - Monday, May 30, 2016 - link
http://www.anandtech.com/show/10375/arm-unveils-bi...YaleZhang - Monday, May 30, 2016 - link
Decreasing the instruction decode width from 3/cycle to 2 to save power is interesting. When you look at the x86 vs RISC debate, everyone tells you complex instruction decoders are a small cost compared to the rest of the core, which is true in terms of area, but no mention about power?So for A73, did ARM reduce the decode width because it was rarely being utilized, or because fundamentally it's power hungry, or both? This seems to suggest why there are no x86 phones. If instruction decode power is a problem for RISC ARM, then it should be even a bigger problem for CISC x86.
StrangerGuy - Monday, May 30, 2016 - link
IIRC, the original K7 only manages to sustain execution on just one decoder out of three, and successive x86 uarches has made decoders more and more capable on per decoder basis, while almost all the x86 IPC gains seems to come from everything except the decoders...I can see the case why ARM moved from 3 to 2 decoders.Krysto - Monday, May 30, 2016 - link
> For example, the A15, A57, A72 all belong to the Austin family of microarchitectures, and as one would have guessed from the name, this is because they originated from ARM's Austin CPU design centre.The A5, A7 and A53 belong to the Cambridge family while the Cortex A12, A17 and today's new A73 belong to the Sophia family, owning its name to the small city of Sophia-Antipolis which houses one of Europe's largest technology parks as well as ARM's French CPU design centre.
So the American ARM CPU cores suck, while the European ones are pretty great. Got it. We should keep this in mind when ARM announces new cores. Stay away from the American ones, adopt the European ones.
kpkp - Monday, May 30, 2016 - link
A72 seems far from bad.psychobriggsy - Monday, May 30, 2016 - link
The American designs use more power and are larger.The Cambridge designs try to use the least power and are small (I expect the A35 is also from here?)
The French designs are larger, but low power.
Each site's designs seem to take about two years, and they're staggered.
The biggest advantage is that if one of the sites really screws up a generation of CPU, it appears the other site will have them covered. If A73 was a turd, nobody would have batted an eyelid at A72 for another year, for example.
djayjp - Monday, May 30, 2016 - link
"...both microarchitectures have trouble in terms of differentiating themselves in terms of performance and power compared to ARM’s own current designs..."You're kidding, right? *cough single threaded performance *cough....
Andrei Frumusanu - Monday, May 30, 2016 - link
I'll cover the topic in more detail in the separate review but neither Qualcomm nor Samsung display an advantage in single-thread performance over current ARM microarchitectures.jjj - Monday, May 30, 2016 - link
Chances are Qualcomm will go FOWLP soon and that's why they want custom A73. On 10nm the die will be tiny and plenty of room to add the DRAM in the same package.StrangerGuy - Monday, May 30, 2016 - link
A72s @ 28nm are already scoring ~1700 Geekbench ST. If A73 on the 28nm adds a conservative 20% puts it in Kyro/Mongoose territory. 14nm A73 would be even stronger still with inherent finfet power efficiency boosts.Not hard to see why custom ARM camp should be really worried, because ARM own designs have the flexibility in process nodes while performance is rapidly catching up with custom.
Shyamkr - Monday, May 30, 2016 - link
Is there any comparative data between A73 and qcom kryo cores?phoenix_rizzen - Monday, May 30, 2016 - link
2016 / 2017 are going to be interesting again in the mobile SoC space.We have the fully custom SoCs from the likes of Qualcomm (Kryo CPU, and Adreno GPU).
The semi-custom SoCs from the likes of Samsung (Mongoose big cores + Cortex-A53 LITTLE cores, and Mali GPU) and Apple (Twister CPU, and PowerVR GPU).
The pure ARM SoCs from the likes of MediaTek, Rockchip, Huewai (sp?), and similar (Cortex-A72/A73 + Cortex-A53/A35 and Mali GPUs).
All on multiple processes (28nm, 16nm, 14nm, high-performance, low-power, FinFETT, etc).
Now we just need someone to combine a high-performance, high-efficiency SoC with a high-efficiency radio and screen (that's not too big) and stick a huge battery inside (4000 mAh should be the new normal; anything under 3000 mAh should be avoided). The era of 2-days-between-charges could be upon us ... so long as the OEMs don't squander the battery savings by making the phones even thinner than they are now!!
2015 may have been boring. But it's certainly not anymore!
aryonoco - Monday, May 30, 2016 - link
I also don't think this will be the last of ARM's big core microprocessor family (A15, A57, A72). As the A73 is purely focused on the consumer market, and that's a good thing. However I don't think ARM is giving up on industrial usage and its dream of breaking into the server market. I expect there will be a successor to the A72, and now that it doesn't have to pretend to want to be used in a mobile phone, it can go bigger and wider. It will probably be named something like Cortex A82 or A92 or something, and it could be an absolute beast in Chromebooks, laptops, and potentially maybe even some micro servers. It would be great competition to Intel's Core m, at a fraction of the price.Fun times to be a microprocessor nerd, fun times!
Thanks Andrei for your great reporting. You are making me almost not miss Anand anymore :-)
radar25 - Monday, May 30, 2016 - link
Great Post, I love this articleradar25 - Monday, May 30, 2016 - link
Great Post, dailyplus3 .wix.com/ I love this articleBwash82 - Monday, May 30, 2016 - link
This seems like an awesome replacement for the A72 but why would you need an A53 now. If two A73s can do more with the same space and power budget as 4 A53s why use the core? Honestly I'm kinda mad this was not the core that was initially pushed as the 64 bit chip. ARM was in a great place with the A17 then they pushed forward with a core that was too big hot and power hungry for mobile devices. Honestly I would love to see what a manufacturer can do with six of these bad boys on a soc. May be overkill but it could be fun.saratoga4 - Tuesday, May 31, 2016 - link
A53 perf/watt is much better than the A72. Take a look at the Kirin 950 results. You can more or less double your efficiency with an A53 vs downclocking an A72.fanofanand - Tuesday, September 20, 2016 - link
He was referencing the A73 not the A72. The A73 has nearly the same efficiency as A53 based on the graph, certainly 2 A73's would use less power than 4 A53's, if the graph is to be believed.CodeJingle - Saturday, June 11, 2016 - link
It has nothing to do with Apple, it is completely obvious jumping from 32-bit to 64-bit means one of the processor generations has to take one for the team.SanX - Tuesday, June 21, 2016 - link
Please use solid or more distinctive colors for curves, with these washed out colors absolutely impossible to understand what is whatGXCoder - Tuesday, July 5, 2016 - link
When will you publish the comparative data between kryo and M1 cores?